Visa wrote in an article that trust has become one of the most valuable assets in today’s digital economy. Every day, millions of payments move seamlessly across borders, powered by a shared confidence that transactions are secure. Yet that confidence has been under strain as fraudsters use artificial intelligence (AI) to sharpen their tactics, turning normal interactions into potential vulnerabilities.
Fraud has always followed the money, but AI has changed its shape and speed. Today, generative tools can write convincing scams in seconds, clone voices from short recordings, and spin up fake websites that mirror legitimate ones. According to Sumsub’s latest Identity Fraud Report published in November 2025, deepfake enabled scams in Singapore have surged by 158 per cent year on year. These AI generated impersonations now make up a significant share of regional fraud attempts, underscoring the need for stronger digital identity verification and prevention measures due to rising sophistication in scams.
From discussions with risk teams, our partners and regulators, it is clear that today’s battleground is not just how much fraud occurs, but how quickly it evolves, and how fast defenders can adapt without eroding the trust they aim to protect.
The AI arms race in payments
The silver lining is that the same technology driving fraud is also securing payments. Each time someone taps a card, buys a coffee, or books a trip online, AI models are quietly at work. Every transaction is risk scored in real-time and assessed against billions of others to flag anomalies. This includes an unfamiliar device, unusual location, or sudden behavioural shift that the consumer does not usually do.
There is an intrinsic balance between security and friction. Declining a legitimate payment creates inconvenience and chips away at the seamless experience consumers expect. To strike the right balance, our partners work with us to combine traditional rule-based systems with machine learning models that continuously retrain themselves.
This multi-layered defence now spans the payment lifecycle: from account creation and device enrolment to authorisation and post transaction monitoring for fraud. With Visa’s recent acquisition of Featurespace, AI systems can now provide that end-to-end comprehensive view, using adaptive behavioural models to risk-score transactions across cards, account-to-account transfers and real-time payments on a single platform. Because this solution is payment-mode agnostic, banks and payment firms can detect new fraud patterns across multiple rails rather than react to new tactics individually. For consumers and businesses, that means suspicious activity can be intercepted earlier, and more good transactions can be approved without friction.
Trust built through partnership
No single institution can see the full threat landscape. That is why collaboration is key and has become the defining strategy for tackling AI-driven fraud. By extension, ensuring trust is the foundational layer that we should continue to keep top-of-mind.
Across Asia Pacific, we are working with different partners to share intelligence on fraud and scams. When a wave of fake investment sites or phishing texts appears in one market, data and insights are quickly passed to partners elsewhere, along with the updating of fraud models to help stop and contain the spread of these new forms of fraud. In addition, we also introduced Visa Scam Disruption practice, a dedicated, proactive capability designed to identify and stop complex scams as they emerge. It combines Visa’s network level data, advanced AI and deep human expertise to detect scam activity early and disrupt it at scale. The practice supports banks, businesses and consumers by strengthening trust and protecting the integrity of the payments ecosystem.
Technology initiatives reflect this shared responsibility. Tokenisation, for instance, replaces card numbers with unique digital tokens that are worthless if stolen, reducing the value of compromised data. Visa estimates that network tokenisation can lower fraud rates by up to 60 per cent and has already prevented hundreds of millions of dollars in attempted fraud globally. Currently, there are approximately 17.5 billion tokens circulating globally, which is more than three times the number of physical cards. We have been collectively working with our issuing and acquiring partners and introducing programmes to encourage tokenisation adoption. The goal is to make online shopping secure for everyone.
In addition to reducing fraud, tokenisation supports the growing use of agentic commerce, where agents can transact on behalf of consumers and businesses—forming a secure foundation for new ways of commerce to grow.
Designing AI around trust
Looking ahead, AI will not just score transactions but influence how people prove who they are and how digital systems act on their behalf. Today’s wallets already use biometrics like fingerprints and facial recognition, while emerging standards such as passkeys aim to reduce dependence on passwords. Soon, trusted agents that are authorised to shop, pay bills, or negotiate for consumers and businesses will become the norm.
This evolution brings convenience but also new challenges. How can merchants confirm whether a request comes from a legitimate agent or a malicious bot? How can banks explain why an AI model blocked one transaction but approved another? And how can regulators ensure oversight of systems that learn and evolve continuously?
The answer lies in making trust a design principle, not an afterthought. That means using AI to strengthen, not erode, public confidence in the financial system.
